Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
  • G06F13/14—Handling requests for interconnection or transfer
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
  • G06F13/14—Handling requests for interconnection or transfer
  • G06F13/16—Handling requests for interconnection or transfer for access to memory bus
  • G06F13/1668—Details of memory controller
  • G06F13/1673—Details of memory controller using buffers
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
  • G06F1/04—Generating or distributing clock signals or signals derived directly therefrom
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures

Definitions

• This invention relates to computer memory systems and, more particularly, to data transfer between a memory controller and a memory unit.
• Computer systems employ many different types of system memory.
• One common type of system memory is implemented using removable memory modules.
• Memory modules come in different types and configurations. However in general, a memory module may be implemented as a printed circuit board having an edge connector and a number of memory devices. The memory module may be plugged into a socket located on a motherboard or other system board.
• a commonly used memory module is known as a dual inline memory module (DIMM), although there are others.
• DIMM dual inline memory module
• memory devices may be nonremovable and may be mounted directly to the motherboard or system board.
• Computer system processor speed and performance have increased rapidly in recent history. However, system memory performance has typically lagged behind. As such, some system performance improvements may be limited by the performance of the system memory. Thus, improvements in system memory bandwidth and capacity may be of great concern to the system architect.
• the memory controller includes an input/output (I/O) circuit including a plurality of input buffers and a plurality of output drivers.
• the I/O circuit may be configured to operate in one of a first mode and a second mode dependent upon a state of a mode selection signal.
• the I/O circuit may be configured to provide a parallel interconnect for connection to one or more memory modules.
• the I/O circuit may be configured to provide a respective serial interconnect for connection to each of one or more buffer units, each configured to buffer memory data that is being read from or written to the one or more memory modules.
• each respective serial interconnect includes a plurality of differential bidirectional data signal paths. Each differential bidirectional data signal path may convey data between a given buffer unit and the memory controller.
• each respective serial interconnect includes a differential command signal path that may convey command information from the memory controller to a given buffer unit.
• each respective serial interconnect includes a plurality of downstream differential unidirectional signal paths and a downstream unidirectional differential clock signal path.
• Each of the downstream differential unidirectional signal paths may convey data, address, and command information from the memory controller to the one or more buffer units.
• the downstream unidirectional differential clock signal path may convey a serial clock signal from the memory controller to each of the one or more buffer units.
• each respective serial interconnect includes a plurality of upstream differential unidirectional signal paths.
• Each of the upstream differential unidirectional signal paths may convey data and cyclic redundancy code (CRC) information from one of the one or more buffer units to the memory controller.
• CRC cyclic redundancy code
• FIG. 1 is a block diagram of one embodiment of a memory system including a high-speed buffer.
• FIG. 2 is a diagram illustrating more detailed aspects of the memory system components of FIG.
• FIG. 3 is a timing diagram illustrating an exemplary burst operation of the embodiments shown in FIG. 1 and FIG. 2.
• FIG. 4 is a flow diagram describing the operation of the embodiments shown in FIG. 1 through
• FIG. 5 is a block diagram of one embodiment of a computer system including the memory system shown in FIG. 1.
• FIG. 6 is a block diagram of an embodiment of the computer system of FIG. 6 including a dual mode memory controller.
• FIG. 7 is a block diagram of one embodiment of a memory system including memory controller having a dual mode interface.
• FIG. 8 a diagram illustrating more detailed aspects of the memory system components of FIG. 7.
• FIG. 9 is a flow diagram describing the operation of the embodiments shown in FIG. 7 and FIG. 8.
• FIG. 10 is a block diagram of one embodiment of a computer system including the memory system shown in FIG. 7.
• FIG. 1 1 is a block diagram of an embodiment of the computer system of FIG. 10 including a dual mode memory controller.
• FIG. 1 1 is a block diagram of an embodiment of the computer system of FIG. 10 including a dual mode memory controller.
• Memory system 10 includes a memory controller 100 coupled to memory units 1 1OA through 1 1OH, and to buffer units 170A through 170J. It is noted that components that include a reference designator having a number and a letter may be referred to by the number only. For example, memory unit 1 1OA may be referred to as memory unit 1 10 where appropriate. It is also noted that memory controller 100 may be a memory controller that is part of a chipset, such as may be used in a Northbridge arrangement. Alternatively, as shown in FIG.
• memory controller 100 may be part of an embedded solution in which memory controller 100 is embedded within a processing node including one or more processor cores, for example.
• memory units 1 lOA-1 1OH may be memory modules such as dual in-line memory modules (DIMM), for example.
• DIMM dual in-line memory modules
• each DIMM may include plurality of memory devices (not shown) such as devices in the dynamic random access memory (DRAM) family of memory devices, for example.
• DRAM dynamic random access memory
• memory units 1 10 of system 10 may be representative of any type of system memory.
• memory controller 100 is coupled to buffer units 170 via a highspeed serial interconnect 160A and 160B.
• each high-speed serial interconnect 160 uses differential signaling techniques.
• High-speed serial interconnect 160 may include a plurality of differential bidirectional data signal paths (DDQ), differential buffer command signal paths (BCMD), differential clock signal paths (WCLK), and differential cyclic redundancy code signal paths (CRC).
• serial interconnect 160A may be used for one channel and is thus coupled to buffer units 170A through 170F
• serial interconnect 160B may be used for the other channel and is coupled to buffer units 170G through 170J. It is noted that in the illustrated embodiment, a portion of each of buffer units 170 E and 17OJ are unused and as such may be used for other purposes, as desired.
• memory controller 100 is coupled to memory units 1 10 via parallel interconnect 165.
• parallel interconnect 165 between memory controller 100 and memory units 1 10 may include address/command signal paths (ADDR/CMD), and clock signal paths (MCLK). Similar to the two serial interconnects shown, there are two ADDR/CMD/MCLK signal paths shown. Each of the ADDR/CMD/MCLK signal paths may be used for a respective memory channel. As shown, one of the ADDR/CMD/MCLK signal paths is coupled to memory units 11OA through 1 10D, and the other
• ADDR/CMD/MCLK signal path is coupled to memory units 1 1OE through 1 1OH. Further, buffer units 170 are also coupled to memory units 1 10 via parallel interconnect 165. As shown, parallel interconnect 165 also includes data paths (DQ) and data strobe signal paths (DQS). In one embodiment, memory controller 100 may control operation of memory units 1 10 by sending addresses and commands via the ADDR/CMD signal paths.
• DQ data paths
• DQS data strobe signal paths
• the DQ data paths may convey data in both directions between the buffer units 170 and memory units 1 10.
• the DQ data paths may include a number of eight-bit (byte-wide) data paths.
• the full data path may be 288 bits wide, but the full data path may be divided into byte-sized portions. It is noted that in one embodiment, the 288 bits may include four check bytes, while in other embodiments, other numbers of check byte may be used. It is also noted that the full data path may include any number of data bits, and be divided into different sized portions.
• the DDQ data paths of the serial interconnects 160 may convey serially the data that was conveyed via the parallel interconnect, and at higher speeds.
• the DDQO signal path may convey data bits corresponding to DQ [0:3]
• the DDQl signal path may convey data bits corresponding to DQ [4:7], and the like, although other mappings are possible.
• buffer units 170 may be part of a single integrated circuit. However, due to the number of pins that may be required for such an implementation, it may be impractical. As such, in one embodiment, the data path may be broken up and grouped into smaller units. Accordingly, in one embodiment, each of buffer units 170 may be a standalone integrated circuit that provides buffer functionality to a respective group.
• each serial buffer unit 170 may serially clock in and store two bytes, and then subsequently transmit those two bytes in parallel, on parallel interconnect 165.
• the serial interconnect 160 may transfer data at four times the rate at which parallel interconnect 165 transfers data on the data signal paths.
• the ADDR/CMD signal paths and the MCLK signal paths may operate at half the rate of data paths of parallel interconnect 165.
• the serial interconnect 160 may transfer data on the DDQ data paths at 6.4GT/S, while the data signal paths DQ/DQS of parallel interconnect 165 may transfer data at 1600MT/S, and the ADDR/CMD and MCLK signal paths may operate at 800MT/s.
• serial buffer units 170 may store any number of bytes before transmitting them onto parallel interface 165. It is also noted that the serial interconnect 160 may be operated at any suitable data rate in relation to parallel interconnect 165.
• CRC signal paths may convey CRC information from each buffer unit 170 to memory controller 100 via a respective unidirectional differential signal path.
• clock signal paths may convey WCLK signals to each of buffer units 170.
• the BCMD signal paths convey buffer commands from the memory controller 100 to each of buffer units 170.
• memory controller 100 may control the operation of buffer units 170 via commands sent via the BCMD signal paths. As such, buffer units 170 may have a normal operation mode and a configuration and test mode.
• memory controller 100 may send read and write commands for both data and pre- and post-ambles, to read and write the data storage, and to adjust the phase offset of the DQ signal paths.
• memory controller 100 may control the configuration, training and testing of the buffer units 170 by sending a variety of loopback commands, CRC control commands, and CRC training pattern commands, for example.
• buffer unit 170 At high data rates the probability of buffer unit 170 or memory controller 100 receiving a bit error is significant. Accordingly, it may be necessary to protect transfers between memory controller 100 and buffer units 170 with an error detection code that will robustly detect multiple bit errors within a protected block.
• a CRC code may be used to provide such multiple bit error detection. More particularly, as shown in FIG. 2, to simplify the logic in the buffer unit and/or the memory modules, and reporting of errors to memory controller 100, buffer unit 170 calculates a CRC based on either the data it is generating or the data that it is receiving. Accordingly, to transfer the CRC information back to memory controller 100, the unidirectional CRC signal paths may be used. As shown in FIG.
• CRC unit 250 may calculate the CRC based on its internal data, and send the CRC data back to memory controller 100.
• memory controller 100 may correct the error by retrying the operation.
• the CRC information may be calculated and sent concurrently with the data on a transfer from buffer unit 170 to memory controller 100 so that the CRC may be available at the same time as the data block it is protecting when it reaches memory controller 100.
• delays associated with calculating the CRC may be mitigated by delays introduced on the data paths during write-to- read, and read-to-write transitions.
• buffer unit 170 may be simplified making this type of control functionality asymmetric. As such, memory controller
• control functionality may dynamically and adaptively adjust the signal characteristics (e.g., phase, etc.) of transmitted write data to enable buffer unit 170 to correctly read the data based upon information received from buffer unit 170.
• memory controller 100 may adjust its internal receiver characteristics to enable memory controller 100 to receive data sent by buffer unit 170.
• memory controller 100 may adjust the phase of clock signals that are provided to buffer unit 170 to enable address and command information to be correctly sampled.
• memory controller 100 may adjust the phase of its transmitted clock and data signals to avoid complex phase shifting circuits in the slave.
• memory controller 100 includes a control unit 101 that is coupled to a transmit unit 102, a receive unit 104, and a clocks unit 106.
• Control unit 101 may calculate phase information based on data received from buffer units 170 that may be used to adjust the phase of various clock edges within memory controller 100. For example, in response to such information as CRC data and read data, control unit
• phase tracking and adjustment circuits shown in FIG. 2
• receive unit 104 receive unit 104
• clocks unit 106 clocks unit 106
• FIG. 2 a diagram illustrating more detailed aspects of the memory system components of FIG. 1 is shown. Components that correspond to those shown in FIG. 1 are numbered identically for clarity and simplicity.
• Memory controller 100 is coupled to serial buffer 170 via a differential serial interconnect 160.
• buffer unit 170 may be representative of any of buffer units 170A through 170J shown in FTG. 1.
• differential serial interconnect 160 includes a differential WCLK signal path, a differential BCMD signal path, a differential CRC signal path, and differential data signal paths DDQ[7:0].
• Memory controller 100 includes a 6.4GHz clock signal, which may be generated by clocks unit 106 of FIG. 1 is coupled to variable phase units 293, 294, 295, and 296, which may be part of clocks unit 106 and may provide the internal clock for memory controller 100.
• the outputs of variable phase units 293, 294, 295, and 296 provide the clock signal for flip-flops (FF) 290, 289, 286, and 284, respectively.
• the variable phase unit 293 is coupled to the clock input of FF 290. Since FF 290 has an inverter 292 coupled in a feedback loop to the input, the 6.4GHz clock is output as a 3.2GHz clock.
• the output of FF 290 is coupled to the input of a differential output driver 291 , the output of which is coupled to the differential WCLK signal path.
• the write data is coupled to the input of FF 286.
• the output of FF 286 is coupled to a differential equalization output driver 287.
• the output of driver 287 is coupled to one signal path of DDQ[7:0].
• DDQ[7:0] For each signal path of DDQ[7:0], a similar write data output path (not shown) may be used.
• one signal path of DDQ[7:0] is coupled to a differential input buffer 283, the output of which is coupled to the input of FF 284.
• the output of FF 284 is provided as read data to other portions (not shown) of memory controller 100.
• the CRC signal path is coupled to a differential input buffer 281, the output of which is coupled to the input of a receiver clock data recovery unit (Rx CDR) 282.
• RxCDR is coupled to per bit offset unit 285, which is coupled to variable phase unit 296.
• Buffer command information is provided to the input of FF 289.
• the output of FF 289 is coupled to a differential equalization output driver 288, which is coupled to the differential BCMD signal path.
• Buffer unit 170 includes a buffer 209 that represents a differential input buffer for each of the DDQ[7:0] signal paths. Buffer 209 is coupled to receive the write data sent on one of the DDQ[7:0] signal paths. The output of buffer 209 is coupled to an input of FF 208.
• the output of FF 208 is coupled to write FIFO 220.
• the output of write FIFO 220 is coupled to a DRAM interface 256, which is representative of the input buffer and output driver circuits used to interface to the memory units 110 via parallel interconnect 165.
• DQS[15:0] there are 16 data strobe signal paths DQS[15:0] and 32 data signal paths DQ[31 :0] as part of parallel interconnect 165.
• the write data from write FIFO may be output to the memory units 1 10 via DQ[31 :0]. It is noted that although only the DQ and DQS signals are shown, other signals have been omitted for simplicity. It is also noted that although not shown as such for simplicity, the MCLK and DQS signals may be differential signals.
• Read data from memory units 110 via DQ[31 :0] is coupled through DRAM interface 256 to one input of a multiplexer (mux) 203.
• the output of mux 203 is provided to an input of FF 206.
• Control logic 255 controls the multiplexer input select of mux 203.
• the output of FF 206 is coupled to a differential equalization data output driver 210 which is coupled to one of the differential signal paths of DDQ[7:0].
• Buffer unit 170 includes control logic 255 which is coupled to receive the buffer command information (BCMD) from memory controller 100 through input buffer 201, which is coupled to the input of FF 202..
• BCMD buffer command information
• the BCMD information may cause control logic 255 to drive write data onto the DQ data paths, or to read data for the DQ data paths, or to enter and exit initialization sequences, etc.
• control logic 255 may control the DRAM interface 256, CRC unit 250, mux 203, as well as other circuits.
• the 3.2GHz clock is coupled to the clock input of FF 202, 205, 208, and 206.
• Each of FF 202, 205, 208, and 206 is shown as a dual edge flip flop, meaning they are configured to latch the 'D' input on both the leading and trailing edge of the input clock signal.
• write data, and BCMD information may be conveyed at 6.4Gb/s on their respective data paths, and input latched using the 3.2GHz clock.
• memory controller 100 operates at 6.4GHz, read data, and CRC information may be conveyed at 6.4Gb/s on their respective signal paths and used within memory controller 100 during certain loop back modes.
• write data when write data is received, it is latched by FF 208 and stored to write FIFO 220.
• Write FIFO 220 may store the data until enough bits are received to be output to the memory units 1 10 via DRAM interface 256.
• memory controller 100 may dynamically and adaptively adjust the signal characteristics (e.g., phase, etc.) of transmitted write data and its internal receiver characteristics, and adjust the phase of the 6.4GHz clock which generates the 3.2GHz clock that is provided to buffer unit 170. More particularly, as mentioned above, receive unit 104 includes sample clock phase adjustment circuits such as RxCDR 282 and offset unit 285 to adjust its own local sample clock phase to more optimally receive data transmitted by buffer unit 170. As such, whenever memory controller 100 is receiving CRC data from buffer unit 170, receive unit 104 may use RxCDR 282, offset unit 285 and variable phase unit 296 to adjust the clock phase of FF 284.
• control unit 101 within memory controller 100 may adjust variable phase unit 293 to adjust the phase of the 6.4GHz clock signal provided to FF 290.
• memory controller 100 may adjust variable phase unit 294 to adjust the phase of the 6.4GHz clock signal provided to FF 289 to allow buffer unit 170 to correctly sample buffer command signals.
• control unit 101 may adjust variable phase unit 295 to adjust the phase of the 6.4GHz clock signal provided to FF 286 to adjust the phase of the write data transmitted to buffer unit 170 to enable buffer unit 170 to more optimally receive the write data.
• FIG. 3 is a timing diagram illustrating exemplary operation of the embodiments shown in FIG. 1 and FIG. 2 during an eight-bit burst. More particularly, the timing diagram shows a 128-byte read/write/read burst.
• the diagram includes the MCLK and ADDR/CMD signals, which are provided by memory controller 100 to memory units 110.
• the diagram also shows the DQ and DQS signals, which convey data and data strobes, respectively, between buffer units 170 and memory units 1 10.
• the remaining signals: DDQ, BCMD, and CRC signal convey information between memory controller 100 and buffer units 170.
• read commands e.g., rdA and rdB
• rdA and rdB were sent to memory units 1 10 by memory controller 100.
• read commands e.g., r ⁇ , rl
• the rdA data appears on the DDQ signal paths.
• the rdA and rdB data is conveyed in parallel from memory units 1 10 to buffer units 170 at twice the MCLK rate (e.g. 1600MT/s).
• the data is conveyed serially from buffer units 170 to memory controller 100 at a much faster data rate (e.g., 6.4GT/S).
• write data may be pre-buffered within buffer unit 170.
• the wrX data and associated BCMD write commands e.g., wl
• the wrX data and associated BCMD write commands are sent to buffer units 170, but that data is not written to memory units 1 10 until later, as indicated by the dotted line.
• the read/write/read sequence may described generally as follows: wrX data is written via the DDQ signal paths to buffer unit 170 by memory controller 100 and stored within buffer unit 170. Memory controller 100 concurrently issues read commands (rdA followed several MCLK cycles later by rdB) to memory units 1 10 via the ADDR/CMD signal paths. Just prior to the rdA data appearing on the DQ bus (e.g., at the end of the wrX data transfer on DDQ), memory controller 100 issues read commands (e.g., r ⁇ , rl ) to buffer unit 170 via BCMD.
• read commands e.g., r ⁇ , rl
• memory controller 100 While the rdA and rdB data is on the DQ bus, memory controller 100 sends write commands (e.g., wrX and wrY) via the ADDR/CMD bus to memory nits 1 10. The rdA and rdB data is latched within buffer unit 170 and sent via DDQ to memory controller 100. Prior to the rdB data transfer on DDQ being complete, memory controller 100 sends write commands (e.g., w ⁇ , w2, and w3) to buffer unit
• the w2 command causes the previously stored wrX data to be written to memory units 1 10 while the w3 write command causes the wrY data that was just sent via the DDQ signal paths to be sent to memory units 1 10 via the DQ data paths.
• memory controller 100 issues a rdC command to memory units 1 10 via the ADDR/CMD signal paths. Some number of cycles later, the rdC data and data strobes appear on the DQ signal paths and DQS signal paths, respectively.
• memory controller 100 issues the read commands (e.g., r0 and rl) to buffer units 170 via the BCMD signal paths, thus enabling buffer units 170 to send the read data via the DDQ data paths.
• the wrZ data is not written to memory units 1 10 during this burst. Instead it is stored within buffer unit 170 for use during the next write burst.
• the CRC is generated and sent to memory controller 170 during read and write operations between memory controller 100 and buffer units 170.
• the CRC is generated from BCMD information, write data, and read data as indicated by the arrows.
• the wl , r ⁇ , w ⁇ commands, the wrX rdA, and rdB data are used to generate the CRC information sent on the CRC signal paths from buffer units 170 to memory controller 170.
• the CRC signal paths may have transitions even when the buffer units 170 are idle (i.e., not transferring data).
• the CRC data drives the RxCDR 282 within memory controller 100. Thus, these transitions enable the read data sample clock to be continuously phase aligned to correctly sample the read data.
• FIG. 4 is a flow diagram describing the operation of the embodiments shown in FIG.
• memory controller 100 may adjust the signal characteristics (e.g., phase, etc.) of transmitted write data to enable buffer unit 170 to correctly read the data based upon information received from buffer unit 170.
• memory controller 100 may adjust its internal receiver characteristics to enable memory controller 100 to correctly receive data sent by buffer unit 170.
• memory controller 100 may adjust the phase of clock signals that are provided to buffer unit 170, and to adjust the phase of the BCMD signal to enable buffer command information to be correctly sampled by buffer units 170.
• control logic 255 causes buffer unit 170 to come out of reset in a training mode (block 405).
• all bidirectional signal path drivers e.g., DDQ, DQ, and DQS
• DDQ, DQ, and DQS may be placed in a high-impedance state (block 410).
• the BCMD signal path is looped back to the CRC signal path during even MCLK cycles (block 415), and a training pattern (e.g., 10101010...) is output on the CRC path during odd MCLK cycles (block 420).
• Memory controller 100 drives a training pattern on the BCMD signal path, which is output on the CRC path during the even MCLK cycles (block 425). Memory controller 100 obtains bit-lock and byte-lock receiving known data patterns on CRC path (block 430). In addition, memory controller 100 adjusts the phase of the BCMD clock signal by adjusting variable phase unit 294, so that buffer unit 170 may obtain bit-lock (i.e., bit alignment) and byte-lock (i.e., byte alignment) on the BCMD signal path (block 435). More particularly, memory controller 1000 may vary (shift) the pattern being sent by one bit time (UI) to ensure that buffer unit 170 is capturing each bit correctly and shifting in the serial bits and capturing entire eight-bit bytes on the correct byte boundary. Memory controller may then send a buffer command to take buffer unit 170 out of training mode (block 440).
• UI bit time
• memory controller 100 sends a training pattern (e.g., random pattern having a lot of transitions) via the DDQ data paths. This pattern is stored within write FIFO 220 (block 445). Memory controller 100 reads back the stored pattern to obtain bit-lock (block 450). Memory controller 100 adjusts the phase of the write data (e.g., by adjusting variable phase unit 295) to obtain a bit error rate of substantially 50%. The 50% transition error rate may be indicative that the write data is being sampled near an edge. Memory controller 100 then adjusts the phase of the write data back by 0.5 UI. Doing so should cause FF 208, for example, to sample the data near the middle of each data bit.
• a training pattern e.g., random pattern having a lot of transitions
• This pattern is stored within write FIFO 220 (block 445).
• Memory controller 100 reads back the stored pattern to obtain bit-lock (block 450).
• Memory controller 100 adjusts the phase of the write data (e.g., by adjusting variable phase unit 29
• This process may be performed for each DDQ signal path (block 455).
• memory controller 100 sends a training pattern via DDQ data paths.
• the training pattern may have a different pattern for each byte.
• memory controller 100 may shift the training pattern data in one UI increments. If the CRC information is correct, byte-lock is established (block 460).
• memory controller 100 attempts to obtain read data byte-lock. In one embodiment, memory controller 100 reads back the byte-locked training pattern (block 465). At this point, the serial interconnect should be aligned such that both bit-lock and byte-lock have been obtained in the write and read directions.
• the parallel DRAM interface 256 may be aligned. More particularly, in one embodiment, memory controller 100 may adjust the WCLK phase while preserving the BCMD and DDQ write phase alignment, until the write phase DQS edges are aligned with the appropriate MCLK edges (block 470). [0052] Once the buffer unit 170 serial and parallel interconnects are aligned, during normal operation memory controller 100 may perform write phase training of the serial interconnect 160 using training patterns as described above. This training may be performed at predetermined intervals. Likewise, during idle periods memory controller 100 may monitor and adjust BCMD and CRC alignment by sending a number of idle commands to buffer unit 170. These idle commands may cause predetermined transition rich CRC patterns to be sent on the CRC signal path (block 475).
• FIG. 5 a block diagram of an exemplary embodiment of a computer system that includes the memory system of FIG. 1 and FIG. 2 is shown. It is noted that components that correspond to components shown in FIG. 1 and FIG.2 are numbered identically for clarity and simplicity.
• Computer system 500 includes a processing node 650 coupled to memory buffers 170 and to memory units 1 10.
• the buffer units 170 may be integrated circuit chips mounted to the motherboard, and the memory units 1 10 may plug into sockets.
• the buffer units 170 may be integrated circuit chips mounted to a daughter board that may plug into a memory daughter card socket.
• the daughter boards may have sockets for which to plug the memory units 1 10 in a riser arrangement.
• the processing node 650 includes processor cores 601 coupled to a memory controller 100. It is noted that there may be any number of processor cores 601 within processing node 650.
• memory controller 100 signals are coupled to memory buffers 170 via a differential serial interconnect 160, and to memory units 170 via a parallel interconnect 165.
• the serial interconnect includes unidirectional CRC signal paths, unidirectional WCLK signal paths, unidirectional BCMD signal paths, and bidirectional data signal paths.
• the parallel interconnect 165 includes bidirectional data and data strobe signal paths between the memory buffers 170 and memory units 1 10.
• parallel interconnect 165 includes unidirectional ADDR/CMD and MCLK signal paths between processing node 650 and memory units 1 10. It is noted that in addition to the ADDR/CMD signals, there may be other signals such as chip select, bank select, and others included on the parallel interconnect 165, however, they have been omitted here for simplicity. It is also noted that although not shown as such for simplicity, the MCLK and DQS signals may be differential signals.
• FIG. 6 a block diagram of one embodiment of a computer system including memory controller having a dual mode memory interconnect is shown.
• Computer system 700 is similar to computer system 500 shown in FIG. 5.
• computer system 700 also includes a processing node 650 coupled to memory buffers 170 and to memory units 1 10.
• memory controller 710 differs from memory controller 100 of FIG. 5 because it is a dual-mode memory controller. More particularly, as described in greater detail below, memory controller 710 may be selectively configured to operate with either a parallel interconnect to memory units 1 10 or a serial interconnect for use with buffer units 170.
• memory controller 710 may be configured to operate in a first mode to provide a parallel memory interconnect that may be compatible with a variety of memory specifications.
• memory units 1 10 may be compatible with DDR2, DDR3, or other specifications as desired.
• memory controller 710 may provide, as its parallel interconnect, a parallel interconnect that is compatible with DDR2, and DDR3 technology, as desired.
• memory controller 710 may be configured to operate in a second mode to provide a differential serial interconnect such as serial interconnect 160 of FIG. 1 and FIG. 2.
• a configuration unit 720 may determine and select the configuration of I/O circuits 71 1 within memory controller 710.
• the mode of memory controller 710 may be selected using a hardwired external pin of processing node 600.
• one or more external select pins of processing node 600 may be hardwired to circuit ground as shown, or to VDD or some other voltage, for example.
• Configuration unit 720 may detect the selection pin state, and then configure I/O circuits 71 1 of memory controller 710 accordingly.
• memory controller mode may be selected during system start-up during execution of BIOS 605 or other system level software.
• BIOS 605 BIOS 605
• I/O circuits 71 1 is a parallel interconnect that includes signal paths such as DQ, DQS, ADDR/CMD, and MCLK, for example.
• the I/O circuits 71 1 changes to a differential serial interconnect that is coupled to memory buffer unit 170 (dotted line) such as shown in FIG. 1 , HG. 2, and HG. 5.
• I/O circuits 71 1 may include a plurality of output drivers and input buffers. Some of the drivers and buffers may be differential circuits and some may be single-ended. In one embodiment, depending on the mode, the connections between the various I/O pins of the processing node and the drivers and buffers may be changed.
• portions of I/O circuits 71 1 may operate like a programmable interconnect.
• the CRC/DQS signal paths may change between bidirectional DQS signal paths and unidirectional CRC signal paths.
• the DQS/BCMD may also change between bidirectional DQS signal paths and unidirectional BCMD signal paths.
• the WCLK/DQS signal paths may change between bidirectional DQS signal paths and unidirectional WCLK signal paths.
• the DDQ/DQ signal paths may change between bidirectional single ended DQS signal paths and bidirectional differential data DDQ signal paths.
• FIG. 7 a block diagram of another embodiment of a memory system including a highspeed buffer is shown.
• the memory system 80 includes a memory controller 800 coupled to memory units 1 1OA through 1 1OH, and to buffer units 870A through 870D.
• memory controller 800 may also be a memory controller that is part of a chipset, such as may be used in a Northbridge arrangement.
• memory controller 800 may be part of an embedded solution in which memory controller 100 is embedded within a processing node including one or more processor cores, for example.
• memory units 1 1 OA-1 1OH may be representative of memory modules such as dual in-line memory modules (DIMM), for example, as described above.
• DIMM dual in-line memory modules
• the memory units may conform to various technologies such as DDR2 and DDR3, for example.
• serial interconnect 860A through 860D each serial interconnect 860 uses differential signaling techniques.
• serial differential interconnects 860A - 86OD may each include an upstream link and a downstream link to each buffer unit 870.
• a downstream link may include a plurality of downstream serial data signal paths (DSD) and a corresponding downstream serial clock signal path (DSCLK) that may be used to clock the data into buffer units 870.
• DSD downstream serial data signal paths
• DSCLK downstream serial clock signal path
• each upstream link may include a plurality of upstream serial data signal paths (USD) and a corresponding upstream serial clock signal path (USCLK) that may be used to clock the data into memory controller 800.
• USD upstream serial data signal paths
• USBK upstream serial clock signal path
• serial interconnect 16OA may be used for one channel and is thus coupled to buffer unit 870A
• serial interconnect 860B may be used for the second channel and is coupled to buffer units 870B
• serial interconnect 860B may be used for the third channel and is coupled to buffer units 870C
• serial interconnect 860D may be used for the fourth channel and is coupled to buffer units 870D.
• serial interconnect 860 uses data signal paths that each convey data, CRC, and ADDR/CMD information.
• serial interconnect 860 may use a packet protocol in which the packets may include encodings to designate whether the payload is ADDR/CMD or data.
• each packet may have a format that has dedicated bit times for CRC information, and payload (e.g., data or ADDR/CMD).
• buffer units 870A -870D are coupled to memory units 1 10 via parallel interconnect 865.
• parallel interconnect 865 may include data paths (DQ), data strobe signal paths (DQS), address/command signal paths (ADDR/CMD), and clock signal paths (MCLK). It is noted that there may be other signals such as chip select, bank select, check bits, and others included on the parallel interconnect 865, however, they have been omitted here for simplicity. It is also noted that parallel interconnect 865 may include four channels. As shown, one of the channels is coupled to memory units 1 1 OA through 1 10D, another is coupled to memory units 1 1OE through 1 1OH, another is coupled to memory units 1 10J through 1 1OM, and another is coupled to memory units 1 ION through 1 1OR.
• DQ data paths
• DQS data strobe signal paths
• ADDR/CMD address/command signal paths
• MCLK clock signal paths
• the DQ data paths may convey data in both directions between the buffer units 870 and memory units 1 10, while the differential data paths of the serial interconnects 860 may convey serially the data that was conveyed via the parallel interconnect, and at higher speeds.
• a given uplink USD[O] or downlink DSD [0] signal path may convey data bits corresponding to DQ [0:3]
• the USD[I] signal path may convey data bits corresponding to DQ [4:7], and the like, although other mappings are possible.
• the serial links may be asymmetric in terms of the number of serial data pins.
• the uplink may have more data signal paths than the downlink, since it is assumed more bandwidth may be consumed by read operations than by write operations.
• each serial interconnect 860 may transfer data at four times the rate at which parallel interconnect 865 transfers data on the data signal paths.
• the ADDR/CMD signal paths and the MCLK signal paths may operate at half the rate of data paths of parallel interconnect 865.
• the serial interconnect 860 may transfer data on the uplink and downlink data paths at 6.4GTVs, while the data signal paths DQ/DQS of parallel interconnect 865 may transfer data at 6.4GTVs
• memory controller 800 may control the operation of buffer units 870 via commands sent on the DSD signal paths.
• buffer units 870 may have a normal operation mode and a configuration and test mode. For example, during normal data operations memory controller 800 may send read and write commands for both data and pre- and post-ambles, to read and write the data storage, and to adjust the phase offset of the DQ signal paths.
• memory controller 800 may control the configuration, training and testing of the buffer units 870 by sending a variety of loopback commands, CRC control commands, and CRC training pattern commands, for example.
• a CRC code may be used to provide such multiple bit error detection. More particularly, as shown in FIG. 2, CRC information may be generated and sent in both the uplink and the downlink.
• memory controller 100 may correct the error by retrying the operation.
• a CRC error detected in the downstream link may be encoded into the upstream CRC.
• memory controller 800 may include control functionality that may dynamically and adaptively adjust the signal characteristics (e.g., phase, etc.) of transmitted write data to enable buffer unit 870 to correctly read the data based upon information received from buffer unit 870.
• memory controller 800 may adjust its internal receiver characteristics to enable memory controller 100 to receive data sent by buffer unit 870.
• memory controller 800 may adjust the phase of clock signals that are provided to buffer unit 870 to enable address and command information to be correctly sampled.
• the uncertainties of delays in the transmission path for different signals in a bus may require a per bit phase adjustment of a sample clock of a receiver of those signals.
• memory controller 800 may adjust the phase of its transmitted clock and data signals to avoid complex phase shifting circuits in the slave.
• memory controller 800 includes a control unit 801 that is coupled to a transmit unit 802, a receive unit 804, and a clocks unit 806.
• Control unit 801 may calculate phase information based on data received from buffer units 870 that may be used to adjust the phase of various clock edges within memory controller 800. For example, in response to such information as CRC data and read data, control unit 801 may control phase tracking and adjustment circuits (shown in FTG. 8) within transmit unit 802, receive unit 804, and clocks unit 806, respectively. This functionality is described in greater detail below in conjunction with the descriptions of FIG. 8 and FIG. 9.
• differential serial interconnect 860 includes a downstream differential serial clock signal path (DSCLK), and downstream differential data signal paths DSD[1 1 :0].
• differential serial interconnect 860 includes an upstream differential serial clock signal path (USCLK), and upstream differential data signal paths USD[19:0]
• Memory controller 800 includes a 6.4GHz clock signal, which may be generated by clocks unit 806 of FIG. 7.
• the 6.4GHz clock is the internal clock for memory controller 800.
• the output of variable phase unit 890 provides the clock signal for flip-flop (FF) 889.
• the 6.4GHz clock is also coupled to lane deskew circuit 881 , and to the clock input of FF 893 to generate the serial clock DSCLK. Since FF 893 has an inverter 892 coupled in a feedback loop to the input, the 6.4GHz clock is divided by two and output as a 3.2GHz serial clock.
• the 3.2 GHz clock is driven differentially by differential output driver 891.
• the write data, ADDR/CMD, and CRC is provided to the input of FF 889.
• the output of FF 889 is coupled to a differential equalization output driver 888.
• the output of driver 888 is coupled to one signal path of DSD[1 1 :0].
• a similar output path (not shown) may be used.
• one signal path of USD[19:0] is coupled to a differential input buffer 885, the output of which is coupled to the input of FF 886.
• the output of FF 886 is coupled to the input of Lane deskew unit 881.
• the output of lane deskew unit 881 is provided as read data and CRC information to other portions (not shown) of memory controller 800.
• the upstream serial clock signal USCLK is coupled to differential input buffer 887, the output of which is coupled to variable phase unit 882. The output of which is coupled to the clock input of FF 886.
• Buffer unit 870 includes a buffer 801 that represents a differential input buffer for each of the DSD[1 1 :0] signal paths.
• Buffer 801 is coupled to receive the write data, ADDR/CMD, and CRC information sent on one of the DSD[1 1 :0] signal paths.
• a similar output path (not shown) may be used.
• the output of buffer 801 is coupled to an input of FF 802.
• the output of FF 802 is coupled to the input of FF803 .
• the output of F803 is coupled to a command buffer 805, CRC unit 806, write FIFO 807, and an output multiplexer (mux) 809.
• the output of write FIFO 807 is coupled to a DRAM interface 256, which is similar to the DRAM interface describe above in conjunction with the description of FIG. 2. As shown, there are four MCLK signals, ADDR/CMD signals, 16 data strobe signal paths DQS[15:0], and 72 data signal paths DQ[71 :0] as part of parallel interconnect 865.
• the write data from write FIFO 807 may be output to the memory units 1 10 via DQ[71 :0]. It is noted that other signals have been omitted for simplicity. It is noted that although not shown as such for simplicity, the MCLK and DQS signals may be differential signals.
• Read data from memory units 1 10 via DQ[71 :0] may be coupled through DRAM interface 856 to one input of a mux 809.
• the output of mux 809 is provided to an input of FF 810.
• Control logic 855 controls the multiplexer input select of mux 809.
• the output of FF 810 is coupled to a differential equalization data output driver 81 1 which is coupled to one of the differential signal paths of USD[19:0].
• Buffer unit 870 also includes control logic 855 which is coupled to receive the command information (CMD) from memory controller 800.
• the CMD information may cause control logic 855 to drive write data onto the DQ data paths, or to read data for the DQ data paths, or to enter and exit initialization, and test sequences, etc.
• CMD command information
• control logic 855 may control the DRAM interface 856, CRC units 806 and 808, mux 809, as well as other circuits.
• the 3.2GHz clock is coupled to the clock input of FF 810 and to the input of differential equalization data output driver 812, the output of which is the upstream serial clock USCLK.
• the 3.2GHz clock signal is also coupled to the Divide by four unit 804, thus providing and internal 800MHz clock domain, which is the MCLK domain.
• the packets received via the DSD[1 1 :0] signal paths may be provided to CMD buffer 805, write FIFO 807, and CRC unit 806 concurrently.
• the CMD buffer 805 and write FIFO 807 may include packet decode logic (not shown) to enable them to capture their respective packets.
• packet decode logic not shown
• the packet may be decoded by write FIFO 807 and the data is stored within write FIFO 807.
• CMD buffer 805 may discard data payload packets.
• Write FIFO 807 may store the write data until enough bits are received to be output to the memory units 1 10 via DRAM interface 856.
• the packet may be decoded by CMD buffer 805 and the CMD information is stored within CMD buffer 805.
• Write FIFO 807 may discard CMD payload packets. Since all packets may include CRC payload, CRC unit 806 receives all packets and extracts the CRC information.
• memory controller 800 may dynamically and adaptively adjust the signal characteristics (e.g., phase, etc.) of transmitted write data and received read data. More particularly, as mentioned above, receive unit 804 includes sample clock phase adjustment circuits such as lane deskew 881 , and variable phase units 890 and 882 to adjust its own local sample clock phase to more optimally receive data transmitted by buffer unit 870.
• receive unit 804 may use lane deskew and variable phase unit 882 to adjust the clock phase of FF 885.
• control unit 801 within memory controller 800 may adjust variable phase unit 890 to adjust the phase of the write data transmitted to buffer unit 870 to enable buffer unit 870 to more optimally receive the write data.
• FIG. 9 is a flow diagram describing exemplary operation of the embodiments shown in FIG. 7 and FIG. 8. More particularly, an initialization and configuration procedure for establishing and maintaining communication between memory controller 800 and buffer units 870 is described.
• memory controller and buffer unit 870 comes out of reset in a training state 1, or Tl .
• Tl state serial interconnect 860 is operated at 400MT/s (block 905).
• Memory controller 800 uses a dead-reckoned 0.5UI offset to send and receive data (block 910).
• memory controller adjusts the offset to be an approximated point halfway across a given bit time.
• Memory controller 800 sends a command to cause buffer unit 870 to exit the Tl state and to enter the T2 state (block 915).
• buffer unit 870 drives a predetermined pattern such as a 101010... pattern on all bit lanes of the USD link.
• Memory controller obtains bit-lock using the known pattern and adjusts the variable phase unit 882, for example (block 920).
• memory controller 800 sends a buffer command to cause buffer unit 870 to exit the T2 state and to enter the T3 state by driving all ones for eight bit times(block 925), for example.
• buffer unit 870 sends a predetermined pattern such as 101010... to memory controller 800 via the USD signal paths on even MCLK cycles (block 930).
• Buffer unit 870 is configured to loop back the downstream data to the upstream USD signal paths on odd MCLK cycles and to send downstream via the DSD signal paths a different pattern than the 101010... pattern (block 935).
• Memory controller 800 obtains byte-lock using the different patterns.
• Memory controller 800 then adjusts the downstream data phase to allow buffer unit 870 to obtain bit-lock and byte lock (block 940).
• memory controller 800 drives all zeros for eight bit times causing buffer unit 870 to exit the T3 state and to enter the normal operational mode block (945), where memory controller 800 may read and write data to memory units 1 10, etc.
• memory controller 800 may adjust the divide-by-four
• MCLK divider 804 within each buffer unit 870 such that all buffer units 870 may be using the same clock edge (phase) (block 950). More particularly, memory controller 800 may send buffer commands to retard the MCLK phase by one or more bit times.
• memory controller 800 may train the upstream and downstream signal paths (block 955) using a periodic training mode. For example, for downstream training memory controller 800 may write a training pattern to write FIFO 807 using a predetermined training phase offset (block 960). Memory controller 800 may then read back the training pattern and calculate an error sign from the transition values of the pattern (block 965). Using the calculated error sign, memory controller 800 may adjust the downstream data phase (block 970). [0086] For upstream training, memory controller 800 may write a training pattern to write FIFO 807 using a normal phase offset (block 975).
• Memory controller 800 may then read back the stored training pattern and calculate an error sign from the transition values of the pattern using another predetermined training phase offset (block 980). Using the calculated error sign, memory controller 800 may adjust the upstream sample phase (block 985). Once the periodic training is complete, buffer unit 870 is placed back into normal mode as describe above in block 945. [0087] Turning to FIG. 10, a block diagram of one embodiment of a computer system including the memory system of FIG. 7 is shown. It is noted that components that correspond to components shown in FIG. 7 and FIG.8 are numbered identically for clarity and simplicity. Computer system 1 100 includes a processing node 1 150 coupled to memory buffers 870 and to memory units 1 10. [0088] Similar to the computer system shown in FIG.
• the buffer units 870 may be integrated circuit chips mounted to the motherboard, and the memory units 1 10 may plug into sockets.
• the buffer units 870 may be integrated circuit chips mounted to a daughter board that may plug into a memory daughter card socket.
• the daughter boards may have sockets for which to plug the memory units 1 10 in a riser arrangement.
• the processing node 1 150 includes processor cores 1 101 coupled to a memory controller 800. It is noted that there may be any number of processor cores 1 101 within processing node 1 150.
• memory controller 800 signals are coupled to memory buffers 870 via a differential serial interconnect 860, and to memory units 1 10 via a parallel interconnect 865.
• the serial interconnect 860 includes unidirectional downstream signal paths, unidirectional downstream clock signal paths, unidirectional upstream signal paths, and unidirectional upstream clock signal paths.
• the parallel interconnect 865 includes bidirectional data and data strobe signal paths between the memory buffers 870 and memory units 1 10.
• parallel interconnect 865 includes unidirectional ADDR/CMD and MCLK signal paths between processing node 600 and memory units 1 10. It is noted that in addition to the ADDR/CMD signals, there may be other signals such as chip select, bank select, and others included on the parallel interconnect 865, however, they have been omitted here for simplicity.
• FIG. 1 1 a block diagram of another embodiment a computer system including a memory controller having a dual-mode memory interconnect is shown.
• Computer system 1200 is similar to computer system 1 100 shown in FIG. 10.
• computer system 1200 also includes a processing node 1250 coupled to memory buffers 870 and to memory units 1 10.
• memory controller 1210 differs from memory controller 800 of FIG. 10 because it is a dual-mode memory controller. More particularly, as described in greater detail below, memory controller 1210 may be selectively configured to operate with either a parallel interconnect 865 for direct connection to memory units 1 10 or a serial interconnect 860 for use with buffer units 870 as described above in conjunction with the description of FIG. 7 and FIG. 8.
• memory controller 1210 of FIG. 1 1 may also selectively operate with either a parallel interconnect for direct connection to memory modules that may be compatible with a variety of memory specifications.
• memory units 1 10 may be compatible with DDR2, DDR3, or other specifications as desired.
• memory controller 1210 may provide, as its parallel interconnect, a parallel interconnect 865 that is compatible with DDR2, and DDR3 technology, as desired.
• memory controller 1210 may also be selectively configured to operate in a second mode to provide a serial differential interconnect such as serial interconnect 860 of FIG. 7 and FIG. 8 for connection to buffer units 870.
• serial differential interconnect such as serial interconnect 860 of FIG. 7 and FIG. 8 for connection to buffer units 870.
• a configuration unit 1220 may determine and select the configuration of I/O circuits 1211 within memory controller 1210.
• the mode of memory controller 1210 may be selected using a hardwired external pin of processing node 1250.
• one or more external select pins of processing node 1250 may be hardwired to circuit ground as shown, or to VDD or some other voltage, for example.
• Configuration unit 1220 may detect the selection pin state, and then configure I/O circuits 1211 of memory controller 1210 accordingly.
• memory controller mode may be selected during system start-up during execution of BIOS 1205 or other system level software. [0093] In the illustrated embodiment, in the first mode memory controller 1210 is coupled directly to memory units 1 10.
• I/O circuits 121 1 provides a parallel interconnect that includes signal paths such as DQ, DQS, ADDR/CMD, and MCLK, among others, for example.
• the I/O circuits 121 1 changes to a differential serial interconnect that is coupled to memory buffer unit 870 (dotted lines) such as shown in FIG. 7, FIG. 8, and FIG. 10.
• I/O circuits 121 1 may include a plurality of output drivers and input buffers. Some of the drivers and buffers may be differential circuits and some may be single-ended. In one embodiment, depending on the mode, the connections between the various I/O pins of the processing node and the drivers and buffers may be changed.
• portions of I/O circuits 121 1 may operate like a programmable interconnect.
• the DSD signal paths may change between unidirectional differential DDS signal paths and bidirectional single-ended DQ signal paths, as desired.
• the USD signal paths may change between unidirectional UDS signal paths and bidirectional single-ended ADDR/CMD signal paths, and/or bidirectional differential DQS signal paths.
• the DSCLK signal paths may also change between a differential unidirectional clock signal path to one or more single-ended MCLK signal paths, etc. It is noted that other pin combinations are possible and contemplated.
• This invention may generally be applicable to microprocessors.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Dram (AREA)
• Static Random-Access Memory (AREA)
• Information Transfer Systems (AREA)

Priority Applications (4)

Applications Claiming Priority (2)

Publications (1)

Family

ID=39175571

Family Applications (1)

Country Status (8)

Cited By (9)

Families Citing this family (30)

Citations (3)

Family Cites Families (19)

• 2006
  • 2006-10-31 US US11/590,286 patent/US7694031B2/en active Active
• 2007
  • 2007-10-29 CN CN2007800407571A patent/CN101583933B/zh active Active
  • 2007-10-29 KR KR1020097011350A patent/KR20090077015A/ko not_active Application Discontinuation
  • 2007-10-29 TW TW096140532A patent/TWI489456B/zh active
  • 2007-10-29 GB GB0908637A patent/GB2458040A/en not_active Withdrawn
  • 2007-10-29 WO PCT/US2007/022814 patent/WO2008054696A1/en active Application Filing
  • 2007-10-29 JP JP2009535286A patent/JP2010508600A/ja active Pending
  • 2007-10-29 DE DE112007002619.7T patent/DE112007002619B4/de active Active
• 2010
  • 2010-04-02 US US12/753,373 patent/US8019907B2/en not_active Expired - Fee Related

Patent Citations (3)

Also Published As

Similar Documents

Legal Events